Data Scientist - Agentic AI

Description

The company builds the agentic AI workflows behind Marvis, Juniper's AI assistant for network operations.

This role sits at the meeting point of data science, generative AI and production engineering. You will design, build and evaluate the reasoning pipelines, tool-calling patterns, skills and MCP server integrations that let Marvis diagnose, troubleshoot and fix complex networking problems on its own.

The ideal candidate has hands-on experience with LLM agent frameworks (LangGraph preferred) and the engineering discipline to ship reliable, observable AI systems in a cloud-native setup.

Level expectations: your work affects technical components of products, solutions or services on an ongoing basis. You are the subject matter expert in agentic AI and applied GenAI, apply advanced knowledge to hard business and technical problems, and advise functional and technical teams. You use significant independent judgment to pick the best approach, and you may lead projects and mentor colleagues.

This role is hybrid, with an average of 2 days per week in an HPE office.

Responsibilities

• Build, refine and iterate on agentic workflows in LangGraph, including ReAct loops, dynamic tool selection and binding, multi-step reasoning and self-correction.
• Create and maintain MCP (Model Context Protocol) servers and skills: define tool schemas, implement domain tools, write SKILL.md playbooks, and manage versioning, deployment and monitoring.
• Integrate LLM features at production scale, covering structured outputs, streaming, function and tool calling, prompt engineering and error handling across agent execution paths.
• Build and tune retrieval and memory services, including RAG pipelines, vector-store semantic search, hybrid retrieval, and long-term semantic, episodic and procedural agent memory.
• Design and run evaluation frameworks for non-deterministic agent systems: define metrics, build test harnesses, run A/B tests on skills and tool setups, and keep improving quality.
• Work with network engineers and product managers to turn troubleshooting playbooks into agentic workflows, skills, tools and data pipelines.
• Write data analysis and transformation code that runs in sandboxed execution environments (Code Mode), including multi-tool orchestration scripts, aggregation and visualization.
• Deploy and run containerized services on Kubernetes, and contribute to CI/CD pipelines, container images, health probes and resource tuning.
• Own observability for agent workflows: tracing, logging, cost tracking and performance monitoring to keep non-deterministic systems reliable in production.

Requirements

• A bachelor's, master's or PhD in computer science, data science, mathematics, statistics or a closely related quantitative field, or in engineering or information systems.
• Typically 4 to 7 years building production ML or AI systems, with at least 1 year working hands-on with generative AI and LLM applications.
• Production experience with agent frameworks such as LangGraph (strongly preferred), LangChain, Claude Agent SDK or equivalent, beyond prototypes.
• Solid grasp of agent design patterns: ReAct loops, tool and function calling, dynamic tool binding, skill-based execution, multi-step planning and self-correction.
• Hands-on work with MCP or similar tool-serving protocols, including tool schema design, server implementation and registry management.
• Experience integrating LLM APIs at scale: prompt engineering, structured outputs, streaming, error handling and cost control.
• Experience designing RAG pipelines: chunking, re-ranking, hybrid search, vector stores (OpenSearch or equivalent) and relevance tuning.
• Experience building evaluation and test frameworks for non-deterministic AI systems, including offline evals, A/B tests and LLM-as-judge.
• Strong base in statistical and machine learning methods such as anomaly detection, time-series analysis, clustering or causal inference.
• Practical judgment on when to use retrieval versus fine-tuning, prompt engineering versus structured generation, and how to debug model behavior in production.
• Proficient Python developer with experience in production codebases, not only notebooks.
• Kubernetes experience: deploying, scaling and managing Deployments, Services, ConfigMaps, Secrets and health probes.
• CI/CD experience for automated build, test and deploy using tools such as Jenkins, GitHub Actions, ArgoCD or similar.
• Container image work with Docker, including building, tagging and versioning, plus familiarity with a registry such as ECR or GCR.
• Backend development with FastAPI or equivalent, and REST or GraphQL API design.
• Observability experience for AI systems with tools such as LangFuse, Prometheus or equivalent.
• Preferred, not required: agent memory systems (LangMem or custom architectures), sandboxed code execution (E2B, Firecracker or similar), networking knowledge (wireless and wired diagnostics, troubleshooting), and AWS Bedrock, OpenSearch Serverless or similar managed AI services.
• Preferred: strong written and verbal communication, including explaining technical designs to senior leadership.

Conditions

Hybrid role with an average of 2 days per week in an HPE office.

Job family: Engineering. Job level: TCP03.

HPE is an equal opportunity employer and makes decisions based on qualifications, merit and business need. HPE will also follow applicable laws on employer use of arrest and conviction records.

Benefits

• Health and wellbeing benefits covering physical, financial and emotional wellbeing.
• Career development programs that support goals such as becoming a knowledge expert or moving into another division.
• Accommodation available during the application and interview process on request through HPE's accessibility form.